Files
securityonion/salt/postgres/tools/sbin/so-telegraf-postgres
T
Mike Reeves 5d36d00dec Fix Telegraf metrics falling into pg_partman default partitions
pg_cron's launcher connects to cron.database_name at postmaster start and is
registered BGW_NEVER_RESTART. On a host upgraded onto an existing /nsm/postgres
volume, init-db.sh never runs, so so_telegraf does not exist when PostgreSQL
starts -- the launcher dies and never retries. Salt then creates the database,
the extension, and the schedule, all of which succeed, but no worker is left to
fire the job. partman.run_maintenance_proc() therefore never runs: partitions
stop being premade after create_parent's initial window and every metric lands
in <parent>_default. Retention never fires either.

That state is self-perpetuating. Once the default partition holds rows for a day
with no child, PostgreSQL cannot create that child at all -- attaching it would
violate the default partition's constraint -- so maintenance aborts on the first
parent it reaches. Fixing the scheduler alone does not recover a stalled grid.

Point cron.database_name at the always-present postgres database and register the
job with cron.schedule_in_database targeting so_telegraf, so the launcher no
longer depends on database creation order. group_role drops any registration left
behind in so_telegraf, and both halves are guarded on the live GUC so applying
postgres.telegraf_users before the postgresql.conf change has restarted the
container skips instead of failing.

Maintenance now runs so_admin.telegraf_maintenance(), which drains stranded rows
out of any default partition before calling partman: expired rows are deleted,
the rest are repartitioned. It runs from the state on every highstate as well as
hourly from pg_cron, so a grid whose worker is dead still recovers on its own.
The routines live in a postgres-owned schema so_telegraf has no rights on, since
pg_cron executes them as postgres.

Existing grids are recovered by a marker-guarded repair state that truncates the
non-empty defaults once per host. The backlog is mostly past retention already
and moving tens of GB just to delete most of it is not worth the WAL.

Also raise premake from 3 to 7, reconciled onto existing parents in the retention
subcommand, so an outage has a week of headroom before anything reaches a
default partition, and add a check subcommand reporting partition age, default
occupancy and last job status.
2026-08-05 17:39:39 -04:00

334 lines
12 KiB
Bash

#!/bin/bash
set -e
# Provision Telegraf state inside the so-postgres container.
# Usage: so-telegraf-postgres <subcommand>
# create_db Ensure the so_telegraf database exists.
# group_role Provision the so_telegraf group role, telegraf/partman schemas,
# pg_partman, the so_admin maintenance routines, and the hourly
# pg_cron maintenance job.
# user Create or update a per-minion login role granted to so_telegraf.
# Env: ROLE_USER, ROLE_PASS.
# retention Reconcile partman retention and premake on telegraf parents.
# Env: RETENTION_DAYS.
# maintenance Drain default partitions and run partman maintenance.
# repair One-time recovery: TRUNCATE non-empty defaults, then maintain.
# check Report partition health. Non-zero if any parent is unhealthy.
#
# Once a default partition holds rows for a day with no child, Postgres cannot
# create that child at all -- attaching it would violate the default's
# constraint -- so maintenance drains defaults before calling partman.
cmd="${1:?subcommand required}"
# A function rather than re-invoking $0, which assumes the script is executable.
run_maintenance() {
docker exec -i so-postgres psql -v ON_ERROR_STOP=1 -U postgres -d so_telegraf <<'EOSQL'
SELECT CASE WHEN to_regproc('so_admin.telegraf_maintenance') IS NOT NULL
THEN 'true' ELSE 'false' END AS has_proc \gset
\if :has_proc
CALL so_admin.telegraf_maintenance();
\else
\echo 'so_admin.telegraf_maintenance() is missing; run so-telegraf-postgres group_role first.'
\endif
EOSQL
}
case "$cmd" in
create_db)
if ! docker exec so-postgres psql -U postgres -tAc \
"SELECT 1 FROM pg_database WHERE datname='so_telegraf'" | grep -q 1; then
docker exec so-postgres psql -v ON_ERROR_STOP=1 -U postgres \
-c "CREATE DATABASE so_telegraf"
fi
;;
group_role)
docker exec -i so-postgres psql -v ON_ERROR_STOP=1 -U postgres -d so_telegraf <<'EOSQL'
DO $$
BEGIN
IF NOT EXISTS (SELECT FROM pg_catalog.pg_roles WHERE rolname = 'so_telegraf') THEN
CREATE ROLE so_telegraf NOLOGIN;
END IF;
END
$$;
GRANT CONNECT ON DATABASE so_telegraf TO so_telegraf;
CREATE SCHEMA IF NOT EXISTS telegraf AUTHORIZATION so_telegraf;
GRANT USAGE, CREATE ON SCHEMA telegraf TO so_telegraf;
CREATE SCHEMA IF NOT EXISTS partman;
CREATE EXTENSION IF NOT EXISTS pg_partman SCHEMA partman;
-- Telegraf (running as so_telegraf) calls partman.create_parent()
-- on first write of each metric, which needs USAGE on the partman
-- schema, EXECUTE on its functions/procedures, and write access to
-- partman.part_config so it can register new partitioned parents.
GRANT USAGE, CREATE ON SCHEMA partman TO so_telegraf;
GRANT SELECT, INSERT, UPDATE, DELETE ON ALL TABLES IN SCHEMA partman TO so_telegraf;
GRANT EXECUTE ON ALL FUNCTIONS IN SCHEMA partman TO so_telegraf;
GRANT EXECUTE ON ALL PROCEDURES IN SCHEMA partman TO so_telegraf;
-- partman creates per-parent template tables (partman.template_*) at
-- runtime; default privileges extend DML/sequence access to them.
ALTER DEFAULT PRIVILEGES IN SCHEMA partman
GRANT SELECT, INSERT, UPDATE, DELETE ON TABLES TO so_telegraf;
ALTER DEFAULT PRIVILEGES IN SCHEMA partman
GRANT USAGE, SELECT, UPDATE ON SEQUENCES TO so_telegraf;
-- pg_cron runs these as postgres, so they must not sit in a schema any
-- Telegraf role can create objects in.
CREATE SCHEMA IF NOT EXISTS so_admin AUTHORIZATION postgres;
REVOKE ALL ON SCHEMA so_admin FROM PUBLIC;
CREATE OR REPLACE PROCEDURE so_admin.telegraf_maintenance()
LANGUAGE plpgsql
AS $proc$
DECLARE
r record;
v_default text;
v_rows bigint;
BEGIN
-- No per-parent EXCEPTION handler: partition_data_proc commits internally,
-- and COMMIT is illegal while a subtransaction is active. A failing parent
-- aborts the run and the next pass retries.
FOR r IN
SELECT parent_table, retention
FROM partman.part_config
WHERE parent_table LIKE 'telegraf.%'
ORDER BY parent_table
LOOP
v_default := format('%I.%I',
split_part(r.parent_table, '.', 1),
split_part(r.parent_table, '.', 2) || '_default');
CONTINUE WHEN to_regclass(v_default) IS NULL;
EXECUTE format('SELECT count(*) FROM %s', v_default) INTO v_rows;
CONTINUE WHEN v_rows = 0;
RAISE WARNING 'so_admin.telegraf_maintenance: % rows stranded in %, draining',
v_rows, v_default;
-- Cheaper to delete expired rows than to repartition and then drop them.
IF r.retention IS NOT NULL THEN
EXECUTE format('DELETE FROM %s WHERE "time" < now() - %L::interval',
v_default, r.retention);
COMMIT;
END IF;
-- Bounded so a large backlog drains across several runs.
CALL partman.partition_data_proc(
p_parent_table := r.parent_table,
p_loop_count := 200,
p_source_table := v_default
);
COMMIT;
END LOOP;
CALL partman.run_maintenance_proc();
END;
$proc$;
CREATE OR REPLACE FUNCTION so_admin.telegraf_partition_status()
RETURNS TABLE (
parent_table text,
oldest_child date,
newest_child date,
days_ahead int,
retention text,
default_rows bigint,
default_size text
)
LANGUAGE plpgsql
AS $func$
DECLARE
r record;
v_default regclass;
BEGIN
FOR r IN
SELECT pc.parent_table AS pt, pc.retention AS ret
FROM partman.part_config pc
WHERE pc.parent_table LIKE 'telegraf.%'
ORDER BY pc.parent_table
LOOP
parent_table := r.pt;
retention := r.ret;
SELECT min(d), max(d) INTO oldest_child, newest_child
FROM (
SELECT to_date(substring(c.relname FROM '_p(\d{8})$'), 'YYYYMMDD') AS d
FROM pg_inherits i
JOIN pg_class c ON c.oid = i.inhrelid
WHERE i.inhparent = r.pt::regclass
AND pg_get_expr(c.relpartbound, c.oid) <> 'DEFAULT'
) s;
days_ahead := newest_child - current_date;
v_default := to_regclass(format('%I.%I',
split_part(r.pt, '.', 1),
split_part(r.pt, '.', 2) || '_default'));
IF v_default IS NULL THEN
default_rows := NULL;
default_size := NULL;
ELSE
EXECUTE format('SELECT count(*) FROM %s', v_default::text) INTO default_rows;
default_size := pg_size_pretty(pg_total_relation_size(v_default));
END IF;
RETURN NEXT;
END LOOP;
END;
$func$;
-- Drop the registration older releases left in so_telegraf.
SELECT CASE
WHEN current_setting('cron.database_name', true) IS DISTINCT FROM current_database()
AND EXISTS (SELECT 1 FROM pg_catalog.pg_extension WHERE extname = 'pg_cron')
THEN 'true' ELSE 'false'
END AS drop_stale_cron \gset
\if :drop_stale_cron
DROP EXTENSION pg_cron CASCADE;
\endif
EOSQL
# Guarded on the live GUC so applying this before the postgresql.conf change
# has restarted the container skips rather than failing.
docker exec -i so-postgres psql -v ON_ERROR_STOP=1 -U postgres -d postgres <<'EOSQL'
SELECT CASE WHEN current_setting('cron.database_name', true) = current_database()
THEN 'true' ELSE 'false' END AS cron_here \gset
\if :cron_here
CREATE EXTENSION IF NOT EXISTS pg_cron;
-- cron.schedule_in_database is idempotent by jobname.
SELECT cron.schedule_in_database(
'telegraf-partman-maintenance',
'17 * * * *',
'CALL so_admin.telegraf_maintenance()',
'so_telegraf'
);
\else
\echo 'pg_cron metadata database is not `postgres` yet; skipping job registration.'
\endif
EOSQL
;;
user)
: "${ROLE_USER:?ROLE_USER is required}"
: "${ROLE_PASS:?ROLE_PASS is required}"
# psql does not substitute :vars inside dollar-quoted strings, so the
# conditional CREATE/ALTER is built outside any DO block and dispatched
# with \gexec. format() handles identifier/literal quoting.
docker exec -i so-postgres psql \
-v ON_ERROR_STOP=1 \
-v role_user="$ROLE_USER" \
-v role_pass="$ROLE_PASS" \
-U postgres -d so_telegraf <<'EOSQL'
-- Keep the password out of postgres.log if this DDL errors.
SET log_min_error_statement = panic;
SELECT format(
CASE WHEN EXISTS (SELECT FROM pg_catalog.pg_roles WHERE rolname = :'role_user')
THEN 'ALTER ROLE %I WITH LOGIN PASSWORD %L'
ELSE 'CREATE ROLE %I WITH LOGIN PASSWORD %L'
END,
:'role_user',
:'role_pass'
) \gexec
GRANT CONNECT ON DATABASE so_telegraf TO :"role_user";
GRANT so_telegraf TO :"role_user";
EOSQL
;;
retention)
: "${RETENTION_DAYS:?RETENTION_DAYS is required}"
# \gset + \if guards against a missing pg_partman without using a DO
# block (psql :var substitution doesn't reach into dollar-quoted code).
# premake is reconciled here because telegraf.conf only applies it to
# parents created from now on.
docker exec -i so-postgres psql \
-v ON_ERROR_STOP=1 \
-v retention_days="$RETENTION_DAYS" \
-U postgres -d so_telegraf <<'EOSQL'
SELECT CASE WHEN EXISTS (SELECT 1 FROM pg_catalog.pg_extension WHERE extname = 'pg_partman')
THEN 'true' ELSE 'false' END AS has_partman \gset
\if :has_partman
UPDATE partman.part_config
SET retention = :'retention_days' || ' days',
retention_keep_table = false,
premake = 7
WHERE parent_table LIKE 'telegraf.%';
\endif
EOSQL
;;
maintenance)
run_maintenance
;;
repair)
# Stranded rows are discarded, not repartitioned: most are already past
# retention, and moving tens of GB just to drop them isn't worth the WAL.
docker exec -i so-postgres psql -v ON_ERROR_STOP=1 -U postgres -d so_telegraf <<'EOSQL'
SELECT CASE WHEN EXISTS (SELECT 1 FROM pg_catalog.pg_extension WHERE extname = 'pg_partman')
THEN 'true' ELSE 'false' END AS has_partman \gset
\if :has_partman
\echo 'Default partitions holding stranded Telegraf metrics:'
SELECT format('%I.%I', n.nspname, c.relname) AS default_partition,
pg_size_pretty(pg_total_relation_size(c.oid)) AS size
FROM partman.part_config pc
JOIN pg_class p ON p.oid = pc.parent_table::regclass
JOIN pg_inherits i ON i.inhparent = p.oid
JOIN pg_class c ON c.oid = i.inhrelid
JOIN pg_namespace n ON n.oid = c.relnamespace
WHERE pc.parent_table LIKE 'telegraf.%'
AND pg_get_expr(c.relpartbound, c.oid) = 'DEFAULT'
AND pg_total_relation_size(c.oid) > 0
ORDER BY pg_total_relation_size(c.oid) DESC;
SELECT format('TRUNCATE TABLE %I.%I', n.nspname, c.relname)
FROM partman.part_config pc
JOIN pg_class p ON p.oid = pc.parent_table::regclass
JOIN pg_inherits i ON i.inhparent = p.oid
JOIN pg_class c ON c.oid = i.inhrelid
JOIN pg_namespace n ON n.oid = c.relnamespace
WHERE pc.parent_table LIKE 'telegraf.%'
AND pg_get_expr(c.relpartbound, c.oid) = 'DEFAULT'
\gexec
\endif
EOSQL
# Premake the window so the next flush has somewhere to land.
run_maintenance
;;
check)
docker exec -i so-postgres psql -U postgres -d so_telegraf <<'EOSQL'
\pset border 2
SELECT * FROM so_admin.telegraf_partition_status();
EOSQL
docker exec -i so-postgres psql -U postgres -d postgres <<'EOSQL'
\pset border 2
SELECT CASE WHEN to_regclass('cron.job_run_details') IS NOT NULL
THEN 'true' ELSE 'false' END AS has_cron \gset
\if :has_cron
SELECT d.status, d.return_message, d.start_time
FROM cron.job_run_details d
JOIN cron.job j ON j.jobid = d.jobid
WHERE j.jobname = 'telegraf-partman-maintenance'
ORDER BY d.start_time DESC
LIMIT 5;
\else
\echo 'pg_cron is not installed in this database.'
\endif
EOSQL
unhealthy=$(docker exec so-postgres psql -U postgres -d so_telegraf -tAc \
"SELECT count(*) FROM so_admin.telegraf_partition_status()
WHERE coalesce(default_rows, 0) > 0 OR coalesce(days_ahead, -1) < 1")
if [ "${unhealthy:-1}" != "0" ]; then
echo "so-telegraf-postgres check: $unhealthy telegraf parent(s) unhealthy" >&2
exit 1
fi
echo "so-telegraf-postgres check: all telegraf parents healthy"
;;
*)
echo "Unknown subcommand: $cmd" >&2
exit 1
;;
esac